Gas terminal velocity from MIRO/Rosetta data using neural network approach

Context. The Microwave Instrument on the Rosetta Orbiter (MIRO) board spacecraft was designed to investigate surface and gas activity of comet 67P/Churyumov-Gerasimenko. MIRO spectroscopic measurements carry information about velocity emanating from nucleus surface. Knowledge terminal H 2 O is valuable for interpretation in situ measurements, calibrating 3D coma simulations, studying physics acceleration. Aims. Using a neural network technique, we aim estimate entire dataset nadir geometry pointings. encoded Doppler shift measured rotational transitions o-H 16 18 even when spectral lines are optically thick with quasi or fully saturated line cores. Methods. Neural networks robust nonlinear algorithms that can be trained recognize underlying input output functional relationships. A training set 2200 non-LTE simulated spectra two computed known cometary atmospheres, varying column density, temperature, expansion profiles. Two four-layer used learn then predict lines, respectively. We also quantify accuracy, stability, uncertainty estimated parameter. Results. Once trained, very effective inverting spectra. process August 2014 July 2016, correlations temporal evolution velocities derived lines. highest obtained higher than those differences evolve time reach 150 m s −1 average around perihelion. discussion provided how explain this peculiar behavior.